The Superorganism and the Machine
My favourite animal is one most people have probably never heard of. A siphonophore is not one creature. It is a colony of specialised bodies called zooids, each built for a single task. One catches prey, one digests, one swims, one reproduces, but none can survive alone. Together, they form a single animal that can be longer than a blue whale, drifting through the dark middle of the ocean. The colony holds because no zooid keeps for itself what the whole produces. It is, to me, one of the most uncompromising cases in nature that survival is a thing done together, or not at all—for a colony, and for anything or anyone else that hopes to last.
For most of history we could only study them dead. Siphonophores are too fragile to bring to the surface. This summer, a research vessel off the coast of Brazil found a way around that: it swept a sheet of red laser light through the animals’ transparent bodies and built 3D scans without touching them. Most deep-sea creatures cannot see the red light and so keep drifting undisturbed. Describing a single species can take decades. This team confirmed thirty-one in a fortnight.¹
I believe this is the real story of technology in conservation, and AI is its sharpest instance. It is about time. Conservation rarely fails because we do not know enough, but it can often fail because we find out too late. A single monitoring project throws off millions of images, and a threatened species cannot wait the months it takes to sort them by hand. In response, some conservationists have opted to teach machines to do the sorting. WWF and Google built Wildlife Insights, whose model can name the animal in a photograph in a fraction of a second. After Australia's Black Summer fires, partners ran more than seven million camera-trap images through it across eight burnt regions, work that would once have taken years.² Among the questions it helped answer quickly: had the greater glider, a small wide-eyed marsupial that glides between trees and had just watched its forest burn, taken to the insulated nest boxes bolted to the surviving trunks, its own hollows gone and centuries from regrowing? The cameras showed gliders in every single box. Every one. That was evidence enough to scale the effort across the fire-affected regions.³ The machine did not replace the ranger who climbed the tree or the ecologist who knew where to look. It gave them back what extinction cannot return: time. And the eyes in that forest were not only the machine's. The cameras were placed and checked by rangers, ecologists, and community volunteers; the machine only sorted what they gathered.
A mechanism worth noticing: these systems are only as good as the people feeding them. Record a blackcap on your morning walk through the free Merlin app, and your five seconds of birdsong joins the same current of citizen data that trains these systems. I'm one of ten million who already do.⁴ That is the technology at its best: distributed, contributed, held in common by the people closest to the life it watches.
But "in common" tends to be the exception, not the rule. When a rainforest is surveyed, the images and the genetic sequences become a dataset, and a dataset has an owner; more often a university or company in the North than the community whose forest was recorded. The places richest in life tend to be the ones with the least means to study it, or to act on what the study finds. At its 2025 Congress in Abu Dhabi, the IUCN moved to write the first rules for AI in conservation, built on data sovereignty, the rights of Indigenous peoples, and, most importantly, the capacity to act in the regions that hold the most life and the fewest resources to protect it.⁵ The question, then, is less who owns the data than who is equipped to use it.
If we get that right, the possibilities are extraordinary. We have built an instrument fine enough to find thirty-one new species in a fortnight, in one of the least reachable places on Earth, without so much as touching them. But a scan is not a rescue. A siphonophore holds together because every part is left able to do its work; none commands the others, and none is spare. We are less separate from that arrangement than we like to think. The people closest to the vanishing life are rarely the ones able to act on it—not for want of care, but because they are deciding under harder constraints, with more urgent claims on the little they have. A whole is only as capable as its least-equipped working part. Starve the parts that do the work, and the whole does less than it could, then less than it must.
Path:OS works at the intersection of governance, systems change, and civic capacity. This piece reflects the analytical perspective of the practice. Empirical claims are sourced; where the evidence thins, the text says so.
Notes & Sources
Schmidt Ocean Institute, midwater expedition aboard R/V Falkor (too), tropical South Atlantic off Brazil, 2026: thirty-one species new to science documented in the expedition's first two weeks, using the non-invasive DeepPIV laser-imaging system developed by the Monterey Bay Aquarium Research Institute (MBARI). Schmidt Ocean Institute press materials; reported in ScienceAlert, "31 New Deep-Sea Species Discovered Off the Coast of Brazil" (June 2026), and Popular Science, "31 alien-like marine species discovered off the coast of Brazil." DeepPIV uses a red laser sheet; most deep-sea species lack sensitivity to red wavelengths.
Wildlife Insights, the AI camera-trap platform used by WWF (a partner in the founding coalition alongside Google, Conservation International, and others). In Australia's "Eyes on Recovery" project, following the 2019–20 Black Summer bushfires, partners processed more than seven million camera-trap images across eight fire-affected regions. WWF, "Wildlife Insights" (worldwildlife.org/projects/wildlife-insights); WWF-Australia, "Eyes on Recovery" (wwf.org.au).
WWF-Australia, "Eyes on Recovery": camera surveys of nest boxes installed for endangered greater gliders in burnt areas recorded glider use in 100% of the boxes across the survey period, providing the evidence to scale the effort more widely. WWF, "How Artificial Intelligence Buys Valuable Time to Protect Wildlife" (worldwildlife.org).
Merlin Bird ID, a free app from the Cornell Lab of Ornithology; user recordings feed the Lab's eBird and Macaulay Library datasets, which train the identification models. Cornell Lab of Ornithology, "The Magic of Merlin" (birds.cornell.edu) — more than ten million people have used the app.
IUCN World Conservation Congress 2025, Abu Dhabi, 9–15 October 2025: Motion 143 mandates the development of a Union-wide policy and guidelines for the ethical and ecologically responsible use of AI in conservation, recognising AI's potential for biodiversity monitoring while naming risks including algorithmic bias, environmental impact, and threats to data sovereignty. IUCN, "IUCN Members adopt landmark motions at World Conservation Congress in Abu Dhabi" (iucn.org, October 2025).